The British Journal of Psychiatry (2011) 199, 15–22. doi: 10.1192/bjp.bp.110.082776

Review article Long-term psychodynamic psychotherapy in complex mental disorders: update of a meta-analysis Falk Leichsenring and Sven Rabung

Background Dose–effect relationship data suggest that short-term treatments terminated. Ten studies with 971 patients were psychotherapy is insufficient for many patients with included. chronic distress or personality disorders (complex mental disorders). Results Between-group effect sizes in favour of LTPP compared with Aims less intensive (lower dose) forms of psychotherapy ranged To examine the comparative efficacy of long-term between 0.44 and 0.68. psychodynamic psychotherapy (LTPP) in complex mental Conclusions disorders. Results suggest that LTPP is superior to less intensive forms of psychotherapy in complex mental disorders. Further Method research on long-term psychotherapy is needed, not only for We conducted a meta-analysis of controlled trials of LTPP psychodynamic psychotherapy, but also for other therapies. fulfilling the following inclusion criteria: therapy lasting for at least a year or 50 sessions; active comparison conditions; Declaration of interest prospective design; reliable and valid outcome measures; None.

From both a clinical and a health-economic perspective it is mental disorders, and such patients report significantly greater important to distinguish between patients who will benefit deficits in social and occupational functioning.9,10 sufficiently from short-term psychotherapy and those for whom Some data suggest that long-term psychotherapy may be long-term psychotherapy is required. Data on dose–effect helpful for these groups of patients.1,3,11–14 This is true not relationships suggest that most patients experiencing acute distress only for psychodynamic therapy, but also for psychotherapeutic benefit from short-term psychotherapy.1 Short-term psycho- approaches that are usually short-term, such as cognitive– therapy may be defined as a treatment of up to 25 sessions;2 behavioural therapy (CBT).14,15 For long-term psychodynamic applying this definition to the data reported by Kopta et al, about psychotherapy (LTPP), however, strong evidence-based support 70% of the patients with acute distress recovered after short-term as yet is lacking. In a recent meta-analysis of the effectiveness of therapy.1 For patients with chronic distress, about 60% recovered LTTP we focused on complex mental disorders which were after 25 sessions. For patients with characterological distress, i.e. defined as personality disorders, chronic mental disorders or personality disorders, the data of Kopta et al suggest that about multiple mental disorders.16 Twenty-three studies fulfilled the 40% recovered after 25 sessions.1 Perry et al estimated the inclusion criteria. Both randomised controlled trials (RCTs) and length of treatment necessary for patients with personality quasi-experimental observational studies were included, allowing disorder to achieve recovery (defined as no longer meeting us to test for differences between study type. As the number of the full criteria for a personality disorder): according to these controlled studies was small, we calculated within-group effect estimates, half of such patients would recover after 1.3 years or sizes throughout. Large and stable effect sizes were reported for 92 sessions, and three-quarters after 2.2 years or about 216 LTPP in patients with these complex disorders.16 For the studies sessions.3 Summing up, the majority of patients with acute including control groups, we compared the within-group effect distress benefit significantly from short-term psychotherapy, sizes between the LTPP conditions and the control conditions: whereas for many patients with chronic distress and for the effect sizes for LTPP were significantly larger than those in the majority of patients with personality disorders, short-term control conditions.16 However, comparing within-group effect psychotherapy seems not to be sufficient. sizes between treatments uses treatment conditions rather than Evidence-based treatments for these groups of patients are studies as units of analysis, which may reduce the effect of particularly important. Personality disorders, for example, are randomisation.17 This may weaken internal validity, but it does not uncommon in both general and clinical populations. They not necessarily imply that internal validity is severely impaired.18 show a high comorbidity with a wide range of Axis I disorders In order to address this problem we decided to update this meta- and are significantly associated with functional impairments.4–6 analysis, including new studies where available. For the Furthermore, personality disorders were found to have a negative comparison of LTPP and the control conditions between-group prognostic impact on depressive disorders.7 For this reason, effect sizes were assessed, focusing on complex mental disorders experts recommend not focusing on the depressive disorder but as defined above. Our 2008 meta-analysis was criticised by some primarily treating the associated personality disorder.7,8 Another authors for addressing an ‘unconventionally broad research population for whom short-term treatment may not be sufficient question’ by including heterogeneous patient populations and are those with multiple mental disorders. A high proportion of comparison conditions.17 On the contrary, however, researchers patients in clinical populations have not just one but several often adopt unnecessarily narrow entry criteria; a broad

15 Leichsenring & Rabung

perspective on meta-analysis covering different patient (c) active treatments applied in the control conditions; populations and settings increases the generalisability and 19 (d) prospective studies of LTPP including pre- and post-treatment usefulness of results. If results are not homogeneous, subgroup or follow-up assessments; analysis can be used to examine the reasons. In the 2008 meta- analysis we carried out several subgroup analyses for different (e) treatments must have been terminated (no study assessing diagnostic groups.16 In line with these considerations, our outcome for ongoing treatments); updated meta-analysis focused on complex mental disorders (f) use of reliable and valid outcome measures; (again defined as personality disorders, chronic mental disorders or multiple mental disorders), addressing the question whether (g) a clearly described sample of patients with ‘complex’ disorders LTPP is superior to shorter or less intensive psychotherapy in (personality disorders, chronic mental disorders or more than treating these disorders. one mental disorder); (h) adult patients (at least 18 years of age); Method (i) sufficient data to allow determination of between-group effect sizes. The procedures followed in our study are consistent with recent We collected studies of LTPP that were published between guidelines for the reporting of meta-analyses.20 January 1960 and April 2010 based on our previous meta-analysis and an updated computerised search of Medline, PsycINFO Definition of LTPP and Current Contents.16 The following search terms were Psychodynamic psychotherapy serves as an umbrella concept used: (psychodynamic OR dynamic OR psychoanalytic* OR encompassing treatments that operate on a continuum of transference-focused OR self psychology OR psychology of self) supportive–interpretive psychotherapeutic interventions.2,21–23 AND (therapy OR psychotherapy OR treatment) AND (study Interpretive interventions aim to enhance patients’ insight into OR studies OR trial*) AND (outcome OR result* OR effect* OR repetitive conflicts sustaining their problems;2 supportive change*) AND (psych* OR mental*) AND (rct* OR control* interventions aim to strengthen abilities that are temporarily OR compar*). In addition, articles and textbooks were manually inaccessible to patients owing to acute stress (e.g. traumatic searched, and we communicated with authors and experts in the events) or have not been sufficiently developed (e.g. impulse field. control in borderline personality disorder). The establishment of a helping (or therapeutic) alliance is regarded as an important Data extraction component of supportive interventions.22 Transference, defined We independently extracted the following information from the as the repetition of past experiences in present interpersonal articles: author names, publication year, psychiatric disorder relations, constitutes another important dimension of the treated with LTPP, age and gender of patients, duration of LTPP, therapeutic relationship. In psychodynamic psychotherapy, number of sessions, type of comparison group, sample size in each transference is regarded as a primary source of understanding group, use of treatment manuals (yes/no), general clinical and therapeutic change.2,22 The emphasis that psychodynamic experience of therapists (years), specific experience with the psychotherapy puts on the relational aspects of transference is a patient group under study (years), specific training of therapists key technical difference from cognitive–behavioural therapies.24 (yes/no), study design (RCT v. effectiveness), duration of follow- The use of more supportive or more interpretive (insight- up period and use of psychotropic medication.16 Disagreements enhancing) interventions depends on the patient’s needs. The were resolved by consensus. Rating was done without masking more severely disturbed a patient is or the more acute the to treatment condition, since evidence suggests that such masking problem, the greater is the need for supportive interventions, is unnecessary for meta-analyses.28 Effect sizes were independently whereas an emphasis on interpretive approaches is more suitable assessed by two raters. Interrater reliability was assessed for the for less disturbed patients.22 Psychodynamic psychotherapy can outcome domains in question: overall outcome, target problems, be carried out either as a short-term (time-limited) or as a psychiatric symptoms, personality functioning and social long-term open-ended treatment. Open-ended psychotherapy in functioning. For all areas interrater reliability was high (r50.95, which treatment duration is not fixed a priori is not identical P40.002).16 to unlimited psychotherapy.22 Short-term treatments are time- limited, usually lasting between 7 and 24 sessions.2 There is no generally accepted standard duration for long-term psycho- Assessment of effect sizes and statistical analysis dynamic psychotherapy. Lamb compiled more than 20 definitions We assessed effect sizes for target problems, psychiatric symptoms, given by experts in the field,25 ranging from a minimum of personality functioning, social functioning and overall outcome. 3 months to a maximum of 20 years. In this meta-analysis we As outcome measures of target problems, we included both included studies that examined psychodynamic psychotherapy patient ratings of target problems and measures referring to the lasting for at least 1 year or 50 sessions. This criterion is consistent symptoms specific to the patient group under study (e.g. measures with the definition given by Crits-Christoph & Barber and other of depression in treatment studies of major depressive disorder or experts in the field.26 a measure of impulsivity for studies examining borderline personality disorder).29 For psychiatric symptoms we included Inclusion criteria and selection of studies both broad measures of psychiatric symptoms such as the We applied the following inclusion criteria, consistent with recent Symptom Check List 90 (SCL-90) and specific measures such as meta-analyses of psychotherapy:27 measures of depression or anxiety.30 For the assessment of personality functioning, measures of personality characteristics (a) studies of psychodynamic therapy meeting the definition were included (e.g. the Millon Clinical Multiaxial Inventory).31 given above;2,21–23 Social functioning was assessed using the Social Adjustment Scale (b) psychodynamic therapy lasting for at least 1 year or at least and similar measures.32 Whenever a study reported multiple 50 sessions; measures for one of the areas of functioning (e.g. target

16 Long-term psychodynamic psychotherapy

psychiatric symptoms), we assessed the effect size for each subtracting the post-treatment mean from the pre-treatment measure separately and calculated the mean effect size of these mean and dividing the difference by the pooled pre-treatment measures within each study. In our previous meta-analysis standard deviation of the measure.41,42 If more than one LTPP outcome measures were assigned either to target problems or to condition or more than one control condition was included, we psychiatric symptoms, personality functioning or social treated them separately in this analysis. Spearman correlations functioning.16,27 In a study of depressive disorders, for example, were assessed between within-group effect sizes and both duration a reduction in depression could be attributed only to target of treatment and number of sessions. problems, not to psychiatric symptoms. However, this procedure may artificially narrow the data basis for the estimation of actual Assessment of study quality therapeutic effects in the respective outcome areas. In order to avoid this problem in this meta-analysis, we first assigned each According to the inclusion criteria described earlier, we analysed outcome measure to one (and only one) of the three domains only prospective studies of LTPP in which reliable outcome of psychiatric symptoms, personality functioning or social measures were used, the patient sample was clearly described functioning. Overall outcome was assessed by averaging the effect and data to calculate effect sizes were reported. In addition, the quality of studies was assessed by use of the scale proposed by sizes of these three areas. To obtain information about changes in 43 target problems, outcome measures referring to criteria specific to Jadad et al. This scale takes into account whether a study is the patient group under study (e.g. measures of depression in described as randomised and double-blind, and whether with- depressive disorders), which were in the first step of evaluation drawals and ‘drop-outs’ are itemised. In psychotherapy research, however, studies cannot be double-blind because the participants assigned to one of the aforementioned three areas, were 16 additionally assigned to the domain of target problems. This know or can easily find out which treatment they receive. Thus, means that the results for target problems are not independent all studies of psychotherapy would have to be given a score of zero of the other three areas, but more realistic estimates of therapeutic on this item of the Jadad scale. Instead of masking of therapists effects will be achieved. As a measure of between-group effect size and patients, the respective requirement in psychotherapy research for continuous measures, we calculated Hedges’ d and the is that any observer-rated outcome measure is rated by assessors associated 95% confidence interval.33 This measure is a variation unaware of the treatment condition. Additionally, the patient of Cohen’s d which corrects for bias due to small sample sizes.33 perspective is of particular importance in psychotherapy. For this Hedges’ d was calculated by subtracting the mean pre-treatment reason, outcome is often assessed by self-report instruments. We to post-treatment or follow-up difference of the control condition therefore decided to give a score of one point on this item if outcome was assessed by masked raters or by reliable self-report from the corresponding difference of LTPP, divided by the pooled 16 pre-treatment standard deviation. This quotient was multiplied by instruments. With this modification, the three items of the Jadad a coefficient J correcting for small sample size to obtain Hedges’ d. scale were independently rated by us for all studies included; a If a study included more than one LTPP or comparison group, we satisfactory interrater reliability was achieved for the total score used the averaged effect sizes of these groups. We aggregated the of the scale (r = 0.92, P50.001). effect sizes estimates (Hedges’ d) across studies, adopting a random effects model which is more appropriate if the aim is to Results make inferences beyond the observed sample of studies.34 To obtain a mean effect sizes estimate we used MetaWin version Ten studies met the inclusion criteria (Fig. 1).11,13,44–51 For three 2.0 for Windows.35 If the data necessary to calculate effect sizes of these studies we received additional information from the were not published in the article, we asked its authors for this authors.13,46,51 Levy et al reported additional data on outcome information. If necessary, signs were reversed so that a positive for the study by Clarkin et al.13,52 In contrast to our 2008 meta- effect size always indicated improvement. In order to examine analysis, we now included the supportive treatment of the study the stability of psychotherapeutic effects, we assessed effect sizes by Clarkin et al as a form of LTPP because of its description by separately for assessments at the termination of therapy and Levy et al as a psychodynamic therapy.13,52 The study by Korner follow-up. If data pertaining to completers and intention-to-treat et al used a non-randomised comparison group.50 Meta-analytic (ITT) samples were reported, the latter were included. To control results, however, have shown that non-randomised comparison for bias related to withdrawal, we additionally carried out group designs yield comparable – if anything, slightly smaller – ITT analyses. For studies that did not report ITT data we effect size estimates to randomised designs.53 For this reason we conservatively set the effects for patients who withdrew after included the study by Korner et al.50 In an RCT by Knekt et al randomisation to zero. By this procedure, the effect sizes reported comparing LTPP, short-term psychodynamic psychotherapy and for the completers sample were adjusted for missing ITT data. If a (short-term) solution-focused therapy in long-standing depressive study, for example, reported a pre–post treatment difference of and anxiety disorders, the authors assessed the effects of the short- 0.40 for a group of 20 patients who completed the study with term treatment groups at predefined time points that did not 5 patients having withdrawn, we used an adjusted difference of exactly represent end of therapy for the short-term treatments.49,54 0.32 (0.40620/25) for the ITT analysis. Tests for heterogeneity Mean duration of treatment was 5.7 months and 7.5 months were carried out using the Q statistic.33 To assess the degree of respectively for these treatments.49 To include the study by Knekt heterogeneity, we calculated the I2 index.36 In cases of significant et al in this meta-analysis, we used the effects of the short-term heterogeneity random effect models are more appropriate.34,37 To treatments assessed after 9 months, which is the time point control for publication bias, tests for asymmetry in funnel plots following most closely the end of the short-term treatments. As and ‘file drawer’ analyses were performed.36–39 Statistical analyses the effect sizes at 9 months were almost identical to those found were conducted using SPSS version 15.0 and MetaWin version at 7 months, no bias was introduced by this procedure. For LTPP 2.0.35,40 Two-tailed tests of significance were carried out for all we used the outcome assessed after 36 months (end of treatment). analyses. The significance level was set to P = 0.05 unless otherwise In another new RCT, Bateman & Fonagy compared LTPP stated. If more is better, outcome should increase with dosage and (mentalisation-based treatment) with a structured clinical duration of treatment. For this analysis we used within-group management approach in the treatment of patients with effect sizes which were calculated for each condition by borderline personality disorder.45 In addition, we received further

17 Leichsenring & Rabung

fail-safe number exceeds 5K + 10, where K is the number of studies.56 For overall outcome the fail-safe number was 66. As this Studies included in 2008 meta-analysis: 23 exceeds 60 (5K + 10), the effect can be regarded as robust. (documented in 29 publications Summing up, we did not find any cogent indication of publication published between 1 January bias. 1960 and 31 May 2008)

Total number of participants Studies excluded: 14 No control group: 7 studies The ten studies included encompassed 466 patients treated with No non-LTPP control group: 5 studies LTPP and 505 patients receiving comparative treatments. Control group unsuitable to calculate between-group effect sizes: 1 study Not study of complex disorders: 1 study Therapy duration For LTPP the mean number of sessions in the ten studies was Potentially relevant studies identified in update of database 120.5 (s.d. = 117.5) and the mean duration of therapy was 78.0 search for trials published weeks (s.d. = 38.2). For the treatments in the control groups the between 1 June 2008 and mean number of sessions was 45.4 (s.d. = 28.1) and the mean 30 April 2010: 2 duration of therapy was 62.9 weeks (s.d. = 24.0).

Study excluded as treatment not terminated at time Follow-up assessments of outcome assessment: 1 Follow-up assessments were carried out in three studies, at intervals of 1–8 years.48,51,57 Studies included in meta-analysis: 10 (documented in 13 publications) Mental disorders The ten controlled studies of complex mental disorders included Fig. 1 Selection of trials for update of authors’ 2008 meta- the treatment of patients with long-standing depressive and 48,49 analysis of long-term psychodynamic psychotherapy (LTPP).16 anxiety disorders (two studies), cluster C personality disorders (one study),51 borderline personality disorder (five studies),11,13,45,47,50 and eating disorders (two studies).44,46 As the number of studies was too small to conduct separate analyses information about another RCT of LTPP which fulfilled the for specific disorders we combined them into one group called inclusion criteria.48 Huber & Klug provided us with data on the ‘complex mental disorders’. comparison groups of their study that were unavailable at the time 16,48 of our previous meta-analysis. Thus, we included this study in Comparison groups this meta-analysis as another RCT. As both the analytic The psychotherapeutic treatments applied in the comparison psychotherapy and the long-term psychodynamic psychotherapy groups included cognitive (behavioural) therapy (CBT/CT; three group of that study fulfilled our criterion for LTPP, we included groups),44,48,51 cognitive analytic therapy (one group),46 both treatments in this category. The ten studies included are dialectical behavioural therapy (DBT; one group),13 short-term described in online Table DS1. psychodynamic psychotherapy (one group),49 solution-focused therapy (one group),49 family therapy (one group),46 structured Tests for publication bias clinical management (one group),45 and routine psychiatric 11,46,47,50 To reduce the ‘file drawer’ effect we tried to identify unpublished treatment as usual (four groups). In addition, one study studies through the internet and by contacting researchers. To test of eating disorders included nutritional counselling as another 44 for publication bias we calculated correlations between sample size control condition. The authors described this condition as not and between-group effect sizes across studies. A significant including psychotherapy. Including nutritional counselling as correlation may indicate a publication bias in which larger one of the control conditions of LTPP might lead to under- effect sizes in one direction are more likely to be published.55 estimating the effects of the control conditions. For this reason Alternatively, the standard error instead of the sample size can we did not include this therapy in the comparison conditions of be used to test for publication bias. Owing to the small number this meta-analysis. Because of the small number of studies of studies providing follow-up data, we assessed these correlations examining one specific comparison treatment, we did not carry only for the post-treatment between-group effect sizes. Since for out separate analyses for the different comparison conditions comparisons with treatment as usual (TAU) smaller sample sizes (e.g. LTPP v. CBT) but combined the treatments into one group (and larger between-group effect sizes) can be expected than called ‘less intensive forms of psychotherapy’. According to this for a comparison with a specific form of psychotherapy, we procedure the question of whether LTPP yielded a better outcome calculated partial correlations in order to control for the type of than less intensive forms of psychotherapy was studied. comparison condition (TAU v. specific psychotherapy). According to the results, the mean partial correlation between outcome and Treatment manuals sample size was r = 0.05 (range 70.06 to 0.14, P40.73); for p Treatment manuals or manual-like guidelines for LTPP were outcome and standard error, r was 0.16 (P40.46). As another 48,49 p applied in all but two studies. test for publication bias we assessed the fail-safe number according to Rosenthal: this is the number of non-significant, unpublished or missing studies that would need to be added to a meta-analysis Tests for heterogeneity in order to change the results of the meta-analysis from significant We used the Q statistic to test for heterogeneity of between-group to non-significant.39 An effect size can be regarded as robust if the effect sizes,33,35 and the I2 index to assess the degree of

18 Long-term psychodynamic psychotherapy

Table 1 Comparison of long-term psychodynamic psychotherapy with other forms of psychotherapy: between-group effect sizes Hedges’ d Outcome domain Number of comparisons da 95% CIb QI2,%

Overall effectiveness 10 0.54 (0.52) 0.26–0.83 11.72 23 Target problems 9 0.49 (0.48) 0.27–0.71 9.12 12 Psychiatric symptoms 9 0.44 (0.41) 0.15–0.73 11.52 31 Personality functioning 7 0.68 (0.63) 0.31–1.04 5.97 0 Social functioning 8 0.62 (0.59) 0.18–1.06 12.44 44

ITT, intention to treat. a. Adjusted for ITT sample. b. Unadjusted d.

heterogeneity (Table 1).36 For Q, all tests of significance yielded are presented for each of the ten studies (Fig. 2). The random insignificant results (P50.09). The I2 index for overall outcome, effects model was applied in order to aggregate effect sizes across target problems, symptoms, personality functioning and social studies: the differences in outcome between LTPP and other forms functioning indicated low to moderate heterogeneity (Table 1).36 of psychotherapy in complex mental disorders were 0.54, 0.49, For follow-up, the number of studies providing data was too 0.44, 0.68 and 0.62 respectively for overall outcome, target limited to calculate reasonable Q and I2 statistics. problems, psychiatric symptoms, personality functioning and social functioning (Table 1). The ITT analysis yielded similar results (Table 1). According to Cohen these effect sizes can be Correlation of quality ratings with outcome regarded as medium to large.41 All between-group effect sizes In order to examine the relationship between study quality and differed significantly from zero (P50.05). Effect sizes can be outcome the between-group effect sizes were correlated with the transformed into percentiles:41 for example, a between-group total score of the Jadad scale for overall outcome, target problems, effect size of 0.54 as identified in overall outcome indicates that general symptoms, personality functioning and social functioning. after treatment with LTPP, patients on average were better off than Owing to the small number of studies providing follow-up data, 70% of the patients treated in the comparison groups. Only three correlations were only calculated for post-treatment assessment studies provided data to assess between-group effect sizes for effect sizes. For this purpose, the average quality score of the follow-up assessments.48,51,57 For this reason the results are only two raters was used. All correlations were non-significant preliminary. For these three studies the between-group effect sizes (P40.14, rs 70.13 to 0.53). Although not statistically significant, were 0.55, 0.54, 0.48, 0.76 and 0.37 respectively for overall the Spearman correlation was relatively high for symptoms outcome, target problems, psychiatric symptoms, personality (r = 0.53). Accordingly, studies of higher quality tended to yield functioning and social functioning. According to these data the larger between-group effect sizes in favour of LTPP for psychiatric differences in favour of LTPP at follow-up are comparable with symptoms. those at the end of treatment.

Effects of LTPP v. other methods of psychotherapy Correlations of outcome with dosage and duration Because of the small number of studies providing data for Including all treatment conditions (LTPP and non-LTPP), all follow-up assessments, between-group effect sizes were only outcome variables except for target problems showed significant assessed for the post-therapy data, except for some preliminary Spearman correlations with the number of sessions (Table 2). analyses. Between-group effect sizes (Hedge’s d) in overall outcome Treatment duration was significantly correlated with improvements

Favours Favours Source Sample size, n Effect size (95% CI) control LTPP

Bachar (1999)44 26 0.58 (70.11 to 1.27) Bateman (1999)11 38 1.76 (1.06 to 2.46) Bateman (2009)45 134 0.65 (0.30 to 1.00) Clarkin (2007)13 62 0.17 (70.27 to 0.61) Dare (2001)46 79 0.21 (70.30 to 0.72) Gregory (2008)47 30 0.70 (70.04 to 1.44) Huber (2006)48 100 0.53 (0.15 to 0.91) Knekt (2008)49 326 0.35 (0.13 to 0.57) Korner (2006)50 60 1.00 (0.46 to 1.54) Svartberg (2004)51 50 0.01 (70.54 to 0.56)

Total 905 0.54 (0.41 to 0.67) ------70.5 0 0.5 1.0 1.5 2.0 2.5 Hedges d (95% CI)

Fig. 2 Comparative effects of long-term psychodynamic psychotherapy (LTPP) on overall outcome (number of patients in analysis sample may differ from intention-to-treat sample).

19 Leichsenring & Rabung

Table 2 Spearman correlations of outcome (pre–post treatment effect sizes) with duration of therapy and number of treatment sessions Overall outcome Target problems Psychiatric symptoms Personality functioning Social functioning

All treatment conditions Duration 0.36* 0.24 0.39* 0.19 0.50* Sessions 0.54** 0.33 0.37* 0.48* 0.63** LTPP only Duration 0.59* 0.28 0.83** 0.18 0.57* Sessions 0.68* 0.28 0.67* 0.31 0.79** Control conditions only Duration 70.10 70.23 70.19 0.30 0.37 Sessions 0.02 70.05 70.19 –a 0.20

LTPP, long-term psychodynamic psychotherapy. a. Insufficient data to calculate correlations. *P50.05, **P50.01 (one-tailed).

in overall outcome, psychiatric symptoms and social functioning. studies would need to be added to this meta-analysis in order to The other correlations were of small to medium size but insignif- change the results of the meta-analysis from significant to non- icant owing to the small number of conditions. Both the direction significant. Furthermore, we found no significant correlation and significance of correlations of outcome with duration or between outcome and sample size nor with standard error of effect dosage of therapy are consistent with the results that showed sizes. We also found no significant correlation between outcome superiority of LTPP over shorter-term treatments. and the methodological quality of the studies as assessed using In some studies treatment lasted for a year or more but the scale proposed by Jadad et al.43 However, the size of some comprised fewer than 50 sessions (online Table DS1). In order correlations may indicate a systematic relationship, in that studies to control for the effect of dosage of LTPP, we additionally assessed of higher quality tended to yield larger between-group effect sizes Spearman correlations between pre–post effect sizes and the in favour of LTPP. Another limitation can be seen in the small number of sessions for the LTPP conditions only (Table 2). Again, number of studies that reported follow-up assessments. It is of all correlations were positive. These correlations were large interest to know whether the between-group effect sizes in favour (40.50) and significant for overall outcome, symptoms and social of LTPP are stable beyond the end of treatment. The results of our functioning. For target problems and personality functioning, previous meta-analysis suggest that the effects of LTPP even small to medium correlations were found that were insignificant. increase after the end of treatment.16 When follow-up data from Thus, the inclusion of studies in the LTPP group in which the the studies included are available, this question can be addressed number of sessions was less than 50 can be assumed to have directly. As another limitation, not all studies reported ITT reduced the effects of LTPP. In the control conditions only, no analyses. In this meta-analysis, however, we could show that significant correlation was found (Table 2). adjusting for missing ITT data did not substantially change the As a further check regarding the importance of dosage, we results. Nonetheless, future studies should include ITT analyses assessed between-group effect sizes without those studies in which whenever possible. fewer than 50 sessions were applied in the LTPP conditions (Dare 44,46,51 et al, Bachar et al, Svartberg et al). For all outcome measures Duration of therapy the effect sizes increased after exclusion of the these three studies (overall outcome from 0.54 to 0.66; target problems from 0.49 to There is no generally accepted standard duration for LTPP. 0.55; psychiatric symptoms from 0.44 to 0.55; personality We included studies that lasted for at least a year or in functioning from 0.68 to 0.77; social functioning from 0.62 to which at least 50 sessions were applied. In some studies treat- 0.72). ment lasted for a year or more but comprised fewer than 50 sessions; for this reason, some of these studies were included in previous meta-analyses as short-term. This was true, for example, Discussion for the study by Svartberg et al in which 40 sessions were applied.27,51 Apparently, the inclusion of studies depends on the A considerable proportion of patients with chronic mental question of research addressed and the specific definition that is disorders or personality disorders do not benefit sufficiently from 1,3 used in a meta-analysis. The correlations between dosage and short-term psychotherapy. Long-term psychotherapy, however, outcome in the LTPP studies reported above suggest that the is associated with higher direct costs than short-term psycho- inclusion of studies in which LTPP lasted for fewer than 50 therapy. For this reason it is important to know whether the sessions reduced the treatment effects of LTPP. However, including effects of long-term psychotherapy exceed those of short-term only studies that fulfilled both the dosage and the duration treatments. In this meta-analysis, LTPP was superior to less criteria would have further reduced the already small number intensive methods of psychotherapy in complex mental disorders. of studies. Future meta-analyses of LTPP or of long-term Furthermore, we found positive correlations between outcome psychotherapy in general should include studies that fulfil and duration or dosage of therapy. Both of these results are 1 both the dosage and the duration criteria. Furthermore, a consistent with data on dose–effect relations. differentiation between long-term, medium-term and short-term One limitation of this meta-analysis may be seen in the therapy might be useful. scarcity of controlled studies. Further studies of LTPP are required to confirm the results and allow for more refined analyses. With a small number of studies it is of particular importance to test for Critical discussion of results publication bias. For that purpose, we applied several measures. This meta-analysis took several points of critique put forward Fail-safe number analysis indicated that for overall outcome, 66 against our 2008 meta-analysis into account, such as lack of

20 Long-term psychodynamic psychotherapy

between-group effect sizes or of ITT analyses, possible publication 17,58 Falk Leichsenring, DSc, Department of Psychosomatics and Psychotherapy, bias or inclusion of inactive control conditions. According to University of Giessen; Sven Rabung, PhD, Department of Medical Psychology, the results presented here we did not find cogent indication for University Medical Centre Hamburg-Eppendorf, Germany any systematic bias. The methodological quality both of our Correspondence: Professor Dr Falk Leichsenring, Department of meta-analyses and of the studies included is comparable to that Psychosomatics and Psychotherapy, University of Giessen, Ludwigstrasse 76, 35392 Giessen, Germany. Email: [email protected] of many studies of CBT.59 Some controlled studies did not meet the inclusion criteria First received 18 May 2010, final revision 6 Nov 2010, accepted 17 Jan 2011 because the majority of patients had not completed their treatment when the effect sizes were assessed. This was true, for example, for the studies by Brockmann et al, Doering et al, Acknowledgements Giesen-Bloo et al and Puschner et al.15,60–62 In the study by Giesen-Bloo, for example, 19 of 42 patients treated with LTPP We thank Dr Martin Rock (Yeshiva University, New York, USA) for information concerning the study by Clarkin et al, and Drs D. Huber and G. Klug and Drs A. Bateman and (45%) were still in treatment when outcome was assessed, and P. Fonagy for giving us access to their data. We also thank John Clarkin, PhD (Weill Medical only 2 patients had completed LTPP; in the comparison group College of Cornell University, New York, USA), Ivan Eisler, PhD (Institute of Psychiatry, King’s College London, UK), Paul Knekt, PhD (National Public Health Institute, Helsinki, Finland) 27 of 44 patients (61%) were still in treatment, and only 6 patients and Martin Svartberg, MD, PhD (Department of Psychiatry, Mount Sinai Hospital, Toronto, 15 had completed the treatment. Data from ongoing treatments Ontario, Canada) for information about their studies. do not provide reliable estimates for treatment outcome at termination or follow-up, for example if patients had received References only half of the ‘dose’ of treatment when outcome was assessed. By analogy, if one runner enters a 100 m race and a second 1 Kopta SM, Howard KI, Lowry JL, Beutler LE. Patterns of symptomatic enters a 10 000 m race, the time taken after 100 m will not be recovery in psychotherapy. J Consult Clin Psychol 1994; 62: 1009–16. representative of the short-distance speed of the second runner. 2 Gabbard GO. Long-term Psychodynamic Psychotherapy: A Basic Text. The runners will adapt their speed to the short or long distance American Psychiatric Publishing, 2004. they are going to face. This is true for patients in psychotherapy 3 Perry J, Banon E, Floriana I. Effectiveness of psychotherapy for personality disorders. Am J Psychiatry 1999; 156: 1312–21. as well.49 Psychotherapy is not a drug that works equally under 4 Grant BF, Hasin DS, Stinson FS, Dawson DA, Chou SP, Ruan WJ, et al. different conditions, but a psychosocial process. Prevalence correlates and disability of personality disorders in the United We compared the effects of LTPP with a group of mixed States. Results from the International Epidemiologic Survey on Alcohol and psychotherapeutic treatments. The control conditions consisted Related Conditions. J Clin Psychiatry 2004; 65: 948–58. of specific forms of psychotherapy, including established forms 5 Kantoja¨ rvi L, Veijola J, Laksy K, Jokelainen J, Herva A, Karvonen JT, et al. such as CBT or DBT, as well as several TAU conditions. Including Co-occurrence of personality disorders with mood, anxiety, and substance abuse disorders in a young adult population. J Pers Disord 2006; 20: 102–12. TAU can be assumed to reduce the mean effect size of the control 6 Lenzenweger MF, Lane MC, Loranger AW, Kessler RC. Personality disorders group; on the other hand, the control conditions included not in the National Comorbidity Survey Replication. Biol Psychiatry 2007; 62: only short-term psychotherapy but also long-term psychotherapy 553–64. applied as long as LTPP in the respective studies (e.g. DBT, CBT), 7 Gunderson JG, Morey LC, Stout RL, Skodol AE, Shea MT, McGlashan TH, et al. in turn increasing the mean effect of the control condition. It is Major depressive disorder and borderline personality disorder revisited: noteworthy that it was on average that duration and the number longitudinal interactions. J Clin Psychiatry 2004; 65: 1049–56. of treatment sessions applied was higher in the LTPP conditions. 8 Elkin I, Shea MT, Watkins JT, Imber SD, Sotsky SM, Collins JF, et al. National Institute of Mental Health Treatment of Depression Collaborative Research Thus, we used the alternative treatments as an unspecific (mixed) Program. General effectiveness of treatments. Arch Gen Psychiatry 1989; control group including both TAU and specific forms of 46: 971–83. alternative psychotherapy. Consequently, we do not claim that 9 Olfson M, Fireman B, Weissman MM, Leon AC, Sheehan DV, Kathol RG, et al. LTPP is superior to any specific form of psychotherapy in complex Mental disorders and disability among patients in a primary care group mental disorders that is carried out equally intensively, rather that practice. Am J Psychiatry 1997; 154: 1734–40. it is superior to less intensive forms of psychotherapeutic 10 Ormel J, VonKorff M, Ustun TB, Pini S, Korten A, Oldehinkel T. Common mental disorders and disability across cultures. Results from the WHO interventions in general. We expect this to be true for other more Collaborative Study on Psychological Problems in General Health Care. intensive approaches of formal psychotherapy as well, for example JAMA 1994; 272: 1741–8. that higher-dose CBT is superior to lower-dose CBT in borderline 11 Bateman A, Fonagy P. The effectiveness of partial hospitalization in the personality disorder. For psychodynamic psychotherapy this treatment of borderline personality disorder, a randomized controlled trial. should also be true. With regard to the hierarchy of evidence, Am J Psychiatry 1999; 156: 1563–9. our comparison of LTPP with a mixed control group including 12 Bateman A, Fonagy P. Treatment of borderline personality disorder with psychoanalytically oriented partial hospitalization: an 18-month follow-up. TAU and specific psychotherapy is stricter than a comparison with Am J Psychiatry 2001; 158: 36–42. a waiting-list group, placebo therapy or pure TAU, but less strict 13 Clarkin JF, Levy KN, Lenzenweger MF, Kernberg OF. Evaluating three (and specific) than a comparison with specific or established treatments for borderline personality disorder: a multiwave study. forms of psychotherapy only.63,64 Am J Psychiatry 2007; 164: 922–8. 14 Linehan MM, Comtois KA, Murray AM, Brown MZ, Gallop RJ, Heard HL, et al. Two-year randomized controlled trial and follow-up of dialectical behavior therapy vs. therapy by experts for suicidal behaviors and borderline personality disorder. Arch Gen Psychiatry 2006; 63: 757–66. Future research 15 Giesen-Bloo J, van Dyck R, Spinhoven P, van Tilburg W, Dirksen C, Without doubt comparisons of LTPP with specific therapies are van Asselt T, et al. Outpatient psychotherapy for borderline personality desirable, both short-term and long-term. At present, however, disorder: randomized trial of schema-focused therapy vs transference- focused psychotherapy. Arch Gen Psychiatry 2006; 63: 649–58. not enough studies are available. For CBT or DBT more 16 Leichsenring F, Rabung S. The effectiveness of long-term psychodynamic comparative studies exist. Thus, it would be interesting to psychotherapy: a meta-analysis. JAMA 2008; 300: 1551–64. compare long-term CBT or DBT with short-term CBT or DBT 17 Kriston L, Ho¨ lzel L, Ha¨ rter M. Analyzing effectiveness of long-term in specific mental disorders. For some mental disorders for which psychodynamic psychotherapy [letter]. JAMA 2009; 301: 930–1. response rates are not satisfactory, such as social anxiety disorder, 18 Leichsenring F, Rabung S. Analyzing effectiveness of long-term 65 experts in the field propose increasing treatment duration. psychodynamic psychotherapy [reply]. JAMA 2009; 301: 932–3.

21 Leichsenring & Rabung

19 Gotzsche PC. Why we need a broad perspective on meta-analysis. BMJ 45 Bateman A, Fonagy P. Randomized controlled trial of outpatient 2000; 21: 585–6. mentalization-based treatment versus structured clinical management for borderline personality disorder. Am J Psychiatry 2009; 166: 1355–64. 20 Moher D, Liberati A, Tetzlaff J, Altman D. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ 2009; 46 Dare C, Eisler I, Russell G, Treasure J, Dodge L. Psychological therapies 339: b2535. for adults with anorexia nervosa. Randomised controlled trial of out-patient treatments. Br J Psychiatry 2001; 178: 216–21. 21 Gunderson JG, Gabbard GO. Making the case for psychoanalytic therapies in the current psychiatric environment. J Am Psychoanal Assoc 1999; 47: 47 Gregory RJ, Chlebowski S, Kang D, Remen AL, Soderberg MG, Stepkovitch J, 679–704. et al. A controlled trial of psychodynamic psychotherapy for co-occurring borderline personality disorder and alcohol use disorder. Psychotherapy 22 Luborsky L. Principles of Psychoanalytic Psychotherapy. Manual for 2008; 45: 28–41. Supportive–Expressive Treatment. Basic Books, 1984. 48 Huber D, Klug G. Munich Psychotherapy Study (MPS): the effectiveness of 23 Wallerstein RS. Psychoanalytic treatments within psychiatry: an expanded psychoanalytic longterm psychotherapy for depression. In Book of Abstracts. view. Arch Gen Psychiatry 2002; 59: 499–500. From Research to Practice (ed Society for Psychotherapy Research): 154. 24 Cutler JL, Goldyne A, Markowitz JC, Devlin MJ, Glick RA. Comparing cognitive Ulmer Textbank, 2006. behavior therapy, interpersonal psychotherapy, and psychodynamic 49 Knekt P, Lindfors O, Harkanen T, Valikoski M, Virtala E, Laaksonen MA, et al. psychotherapy. Am J Psychiatry 2004; 161: 1567–73. Randomized trial on the effectiveness of long-and short-term psychodynamic 25 Lamb WK. A meta-analysis of outcome studies in long-term psychodynamic psychotherapy and solution-focused therapy on psychiatric symptoms during psychotherapy and psychoanalysis. Dissertation. ProQuest, 2004. a 3-year follow-up. Psychol Med 2008; 38: 689–703. 26 Crits-Christoph P, Barber JP. Long-term psychotherapy. In Handbook of 50 Korner A, Gerull F, Meares R, Stevenson J. Borderline personality disorder Psychological Change: Psychotherapy Processes and Practices for the treated with the conversational model: a replication study. Compr Psychiatry 21st Century (eds RE Ingram, CR Snyder): 455–73. Wiley, 2000. 2006; 47: 406–11. 51 Svartberg M, Stiles T, Seltzer MH. Randomized, controlled trial of the 27 Leichsenring F, Rabung S, Leibing E. The efficacy of short-term effectiveness of short-term dynamic psychotherapy and cognitive therapy psychodynamic psychotherapy in specific psychiatric disorders: for Cluster C personality disorders. Am J Psychiatry 2004; 161: 810–7. a meta-analysis. Arch Gen Psychiatry 2004; 61: 1208–16. 52 Levy KN, Meehan KB, Kelly KM, Reynoso JS, Weber M, Clarkin JF, et al. 28 Berlin JA. Does blinding of readers affect the results of meta-analyses? Change in attachment patterns and reflective function in a randomized Lancet 1997; 350: 185–6. control trial of transference-focused psychotherapy for borderline personality 29 Battle CC, Imber SD, Hoehn-Saric R, Nash ER, Frank JD. Target complaints disorder. J Consult Clin Psychol 2006; 74: 1027–40. as criteria of improvement. Am J Psychother 1966; 20: 184–92. 53 Lipsey MW, Wilson DB. The efficacy of psychological, educational, and 30 Derogatis L. The SCL–90 Manual I: Scoring, Administration and Procedures behavioral treatment. Confirmation from meta-analysis. Am Psychol 1993; for the SCL–90–R. Clinical Psychometric Research, 1977. 48: 1181–209. 31 Millon T. Millon Clinical Multiaxial Inventory (3rd edn). National Computer 54 Knekt P, Lindfors O, Laaksonen MA, Raitasalo R, Haaramo P, Ja¨ rvikoski A, Services, 1984. et al. Effectiveness of short-term and long-term psychotherapy on work ability and functional capacity – a randomized clinical trial on depressive and 32 Weissman MM, Bothwell S. Assessment of social adjustment by patient anxiety disorders. J Affect Disord 2008; 107: 95–106. self-report. Arch Gen Psychiatry 1976; 33: 1111–5. 55 Begg CB. Publication bias. In The Handbook of Research Synthesis 33 Hedges LV, Olkin I. Statistical Methods for Meta-analysis. Academic Press, (eds H Cooper, LV Hedges): 399–409. Russell Sage Foundation, 1995. 1985. 56 Rosenthal R. Meta-analytic Procedures for Social Research: Applied Social 34 Hedges LV, Vevea JL. Fixed- and random-effects models in meta-analysis. Research Methods. Sage, 1991. Psychol Methods 1998; 3: 486–504. 57 Bateman A, Fonagy P. 8-year follow-up of patients treated for borderline 35 Rosenberg MS, Adams DC, Gurevitch J. MetaWin: Statistical Software for personality disorder: mentalization-based treatment versus treatment as Meta-analysis, version 2.0. Sinauer, 1999. usual. Am J Psychiatry 2008; 165: 631–8. 36 Huedo-Medina TB, Sanchez-Meca J, Botella J, Marin-Martinez F. Assessing 58 Beck AT, Bhar SS. Analyzing effectiveness of long-term psychodynamic hetereogeneity in meta-analysis. Q statistic or I2 index. Psychol Methods psychotherapy [letter]. JAMA 2009; 301: 931. 2006; 11: 193–206. 59 Leichsenring F, Rabung S. Double standards in psychotherapy research. 37 Quintana SM, Minami T. Guidelines for meta-analyses of counseling Psychother Psychosom 2011; 80: 48–57. psychology research. Couns Psychol 2006; 34: 839–77. 60 Brockmann J, Schlu¨ ter T, Eckert J. Langzeitwirkungen psychoanalytischer und 38 Begg CB, Mazumdar M. Operating characteristics of a rank correlation test verhaltenstherapeutischer Langzeittherapien [Long-term outcome of long- for publication bias. Biometrics 1994; 50: 1088–101. term psychoanalytic and behavioral long-term therapy]. Psychotherapeut 2006; 51: 15–25. 39 Rosenthal R. The file drawer problem and tolerance for null results. Psychol Bull 1979; 86: 638–41. 61 Puschner B, Kraft S, Ka¨ chele H, Kordy H. Course of improvement over 2 years in psychoanalytic and psychodynamic outpatient psychotherapy. Psychol 40 SPSS. SPSS 15.0 Command Syntax Reference. SPSS, 2006. Psychother 2007; 80: 51–68. 41 Cohen J. Statistical Power Analysis for the Behavioral Sciences. Erlbaum, 62 Doering S, Ho¨ rz S, Rentrop M, Fischer-Kern M, Schuster P, Benecke C, 1988. et al. Transference-focused psychotherapy v. treatment by community 42 Hedges LV. Distribution theory for Glass’s estimator for effect size and psychotherapists for borderline personality disorder: randomised controlled related estimators. J Educ Stat 1981; 6: 107–28. trial. Br J Psychiatry 2010; 196: 389–95. 43 Jadad AR, Moore RA, Carroll D, Jenkinson C, Reynolds DJ, Gavaghan DJ, et al. 63 Chambles DL, Hollon SD. Defining empirically supported therapies. Assessing the quality of reports of randomized clinical trials: is blinding J Consult Clin Psychol 1998; 66: 7–18. necessary? Control Clin Trials 1996; 17: 1–12. 64 Gabbard GO, Gunderson JG, Fonagy P. The place of psychoanalytic 44 Bachar E, Latzer Y, Kreitler S, Berry EM. Empirical comparison of two treatments within psychiatry. Arch Gen Psychiatry 2002; 59: 505–10. psychological therapies. Self psychology and cognitive orientation in the 65 Zaider T, Heimberg RG. Non-pharmacologic treatments for social anxiety treatment of anorexia and bulimia. J Psychother Pract Res 1999; 8: 115–28. disorder. Acta Psychiatr Scand Suppl 2003; 417:72–84.

22 The British Journal of Psychiatry (2011) 199, 15–22. doi: 10.1192/bjp.bp.110.082776

Data supplement Yes Yes Yes Yes STPP: 18.5 sessions, 5.7SFT: months 9.8 sessions, 7.5 months 18 months Psychiatric review twice perplus month in-patient admission (90%)visits plus every 2 weeks by CPN Once-weekly individual and group sessions and available telephone consultation Duration RCT CAT: 12.9 sessions, 7 months FT: 13.6 sessions, 1 year TAU: 10.9 sessions, 1 year = 22) = 63) 84 sessions, 18 months Yes n = 30) 12 months = 41) 44 sessions, 26 months Yes n n n Treatment comparison group = 25) 40 sessions, 18.3 months Yes = 17) 12 months Yes n n a ) n = 101) = 15)= 31) 88.7 sessions, 12–18 months Yes 12 months No = 19) = 22) = 97) n n n n n = 22) n n SFT ( Cognitive therapy ( Psychiatric treatment as usual:treatment in-patient (90% of patients)hospitalisation plus partial (72%) plus standard psychiatric aftercare (100%) (TAU, Dialectical behavioural therapy ( FT ( TAU ( Cognitive–behavioural therapy ( ised controlled trial; SFT, solution-focused therapy; STPP, short-term psychodynamic psychotherapy; LTPP Duration Treatment ( Once-weekly 50 min sessions Once-weekly individual therapy plus thrice-weekly group therapy plus once-a-week expressive therapy plus weekly community meeting 12 months LTPP1: two individual weeklyLTPP2: sessions one weekly sessionsessions plus additional as needed LTPP1: 234 sessions, 39LTPP2: months 88 sessions, 34 months a n LTPP2: 30 LTPP1: 43 LTPP2: 35 Eating disordersBorderline personality disorder 22Borderline personality disorder 17Borderline personality disorder 18 months 71 12 months LTPP1: 30 Anorexia nervosa 92 sessions, 18 monthsBorderline personality disorderDepressive disorders (major depressive disorder, recurrent depressive episode or double depression) 15Depressive and Structured 21 anxiety clinical disorders management ( Borderline personality 57.5 disorder sessions, 12–18 months 24.9Cluster sessions, C 1 personality year disorders 128 232 29 sessions, 31.3 months TAU ( 25 12 months 40 CAT sessions, ( 16.9 months STPP ( Cognitive therapy ( TAU ( 51 47 44 13 50 48 49,54 46 52 (2004) Studies of long-term psychodynamic psychotherapy (2008) (1999) (2007) (2006) 11,12 (2006) (2008) (2001) (2006) et al et al et al et al et al et al et al 45 et al et al Table DS1 Levy CAT, cognitive analytic therapy;TAU, CPN, treatment community as psychiatric usual. nurse;a. FT, Intention-to-treat family samples. therapy; LTPP, long-term psychodynamic psychotherapy; RCT, random Clarkin Huber Svartberg Study Mental disorder Patients, Knekt Korner Bateman & Fonagy (2009) Dare Bateman & Fonagy (1999, 2001) Bachar Gregory

1 Long-term psychodynamic psychotherapy in complex mental disorders: update of a meta-analysis Falk Leichsenring and Sven Rabung BJP 2011, 199:15-22. Access the most recent version at DOI: 10.1192/bjp.bp.110.082776

Supplementary Supplementary material can be found at: Material http://bjp.rcpsych.org/content/suppl/2011/06/08/199.1.15.DC1.html References This article cites 52 articles, 4 of which you can access for free at: http://bjp.rcpsych.org/content/199/1/15#BIBL Reprints/ To obtain reprints or permission to reproduce material from this paper, please permissions write to [email protected] You can respond /letters/submit/bjprcpsych;199/1/15 to this article at Downloaded http://bjp.rcpsych.org/ on October 16, 2015 from Published by The Royal College of Psychiatrists

To subscribe to The British Journal of Psychiatry go to: http://bjp.rcpsych.org/site/subscriptions/